Machine learning applied to pore-space geometry in sandstones: a tool for evaluating grain-scale similarity?
Open Access
- 29 September 2021
- journal article
- research article
- Published by Geological Society of London in Quarterly Journal of Engineering Geology and Hydrogeology
- Vol. 55 (1)
- https://doi.org/10.1144/qjegh2020-183
Abstract
The ability to identify similar sandstones to a given sample is important where the provenance of the sample is unknown or the quarry of origin is no longer in operation. In the case of building stones from heritage buildings in protected areas, it may be mandatory Here, a proof of concept for an automated similarity measure is presented by means of a convolutional autoencoder that is able to extract features from a sample thin section and use these features to identify the most similar sample in an existing image library. The approach considers only the shape of the pore space between grains, as, if the pore space alone contains enough information to distinguish between samples, the required image pre-processing and training of a model is greatly simplified. The trained model is able to predict correctly the progenitor quarry of a thin section, from an eight-class dataset of Scottish sandstones, with an accuracy of 47.9%. This prototype, although insufficient for commercial purposes, forms a benchmark for future models against which improvements can be assessed and some of which are suggested.Keywords
Funding Information
- British Geological Survey (N/A)
This publication has 14 references indexed in Scilit:
- Automatic Analysis Of Petrographic Thin Section Images Of SandstonePublished by EAGE Publications bv ,2018
- Biased Dropout and Crossmap Dropout: Learning towards effective Dropout regularization in convolutional neural networkNeural Networks, 2018
- Quantifying Porosity through Automated Image Collection and Batch Image Processing: Case Study of Three Carbonates and an Aragonite Cemented SandstoneGeosciences, 2017
- Rock images classification by using deep convolution neural networkJournal of Physics: Conference Series, 2017
- Deep Residual Learning for Image RecognitionPublished by Institute of Electrical and Electronics Engineers (IEEE) ,2016
- Convolutional neural networks at constrained time costPublished by Institute of Electrical and Electronics Engineers (IEEE) ,2015
- Deep learning in neural networks: An overviewNeural Networks, 2015
- Effects of manual threshold setting on image analysis results of a sandstone sample structural characterization by X-ray microtomographyApplied Radiation and Isotopes, 2012
- Getting to Know Your DataPublished by Elsevier BV ,2012
- Application of Image Analysis to Observe Microstructure in Sandstone and GraniteResource Geology, 2001